4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #include <linux/obd.h>
42 #include <lustre/lustre_idl.h>
43 #include <lustre_lib.h>
44 #include <libcfs/bitmap.h>
45 #ifdef HAVE_SERVER_SUPPORT
46 # include <lu_target.h>
47 # include <obd_target.h>
50 #include <lustre_export.h>
51 #include <lustre_fid.h>
52 #include <lustre_fld.h>
53 #include <lustre_handles.h>
54 #include <lustre_intent.h>
55 #include <lustre_capa.h>
58 #define MAX_OBD_DEVICES 8192
66 struct lov_oinfo { /* per-stripe data structure */
67 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
68 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
69 int loi_ost_gen; /* generation of this loi_ost_idx */
71 unsigned long loi_kms_valid:1;
72 __u64 loi_kms; /* known minimum size */
73 struct ost_lvb loi_lvb;
74 struct osc_async_rc loi_ar;
77 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
80 oinfo->loi_kms_valid = 1;
83 static inline void loi_init(struct lov_oinfo *loi)
87 /* Until such time as we get_info the per-stripe maximum from the OST,
88 * we define this to be 2T - 4k, which is the ext3 maxbytes. */
89 #define LUSTRE_STRIPE_MAXBYTES 0x1fffffff000ULL
91 struct lov_stripe_md {
94 pid_t lsm_lock_owner; /* debugging */
96 /* maximum possible file size, might change as OSTs status changes,
97 * e.g. disconnected, deactivated */
100 /* Public members. */
101 struct ost_id lw_object_oi; /* lov object id/seq */
103 /* LOV-private members start here -- only for use in lov/. */
105 __u32 lw_stripe_size; /* size of the stripe */
106 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
107 __u16 lw_stripe_count; /* number of objects being striped over */
108 __u16 lw_layout_gen; /* generation of the layout */
109 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
112 struct lov_oinfo *lsm_oinfo[0];
115 #define lsm_oi lsm_wire.lw_object_oi
116 #define lsm_magic lsm_wire.lw_magic
117 #define lsm_layout_gen lsm_wire.lw_layout_gen
118 #define lsm_stripe_size lsm_wire.lw_stripe_size
119 #define lsm_pattern lsm_wire.lw_pattern
120 #define lsm_stripe_count lsm_wire.lw_stripe_count
121 #define lsm_pool_name lsm_wire.lw_pool_name
123 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
125 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
128 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
132 if (lsm_is_released(lsm))
137 static inline int lov_stripe_md_size(unsigned int stripe_count)
139 struct lov_stripe_md lsm;
141 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
146 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
148 /* obd info for a particular level (lov, osc). */
150 /* Lock policy. It keeps an extent which is specific for a particular
151 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
152 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
153 ldlm_policy_data_t oi_policy;
154 /* Flags used for set request specific flags:
155 - while lock handling, the flags obtained on the enqueue
156 request are set here.
157 - while stats, the flags used for control delay/resend.
158 - while setattr, the flags used for distinguish punch operation
161 /* lsm data specific for every OSC. */
162 struct lov_stripe_md *oi_md;
163 /* obdo data specific for every OSC, if needed at all. */
165 /* statfs data specific for every OSC, if needed at all. */
166 struct obd_statfs *oi_osfs;
167 /* An update callback which is called to update some data on upper
168 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
169 * request in osc level for enqueue requests. It is also possible to
170 * update some caller data from LOV layer if needed. */
171 obd_enqueue_update_f oi_cb_up;
172 /* oss capability, its type is obd_capa in client to avoid copy.
173 * in contrary its type is lustre_capa in OSS. */
178 struct list_head typ_chain;
179 struct obd_ops *typ_dt_ops;
180 struct md_ops *typ_md_ops;
181 struct proc_dir_entry *typ_procroot;
182 struct proc_dir_entry *typ_procsym;
183 __u32 typ_sym_filter;
186 struct lu_device_type *typ_lu;
187 spinlock_t obd_type_lock;
197 struct timeout_item {
198 enum timeout_event ti_event;
199 cfs_time_t ti_timeout;
202 struct list_head ti_obd_list;
203 struct list_head ti_chain;
206 #define OBD_MAX_RIF_DEFAULT 8
207 #define OBD_MAX_RIF_MAX 512
208 #define OSC_MAX_RIF_MAX 256
209 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
210 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
211 #define OSC_DEFAULT_RESENDS 10
213 /* possible values for fo_sync_lock_cancel */
215 NEVER_SYNC_ON_CANCEL = 0,
216 BLOCKING_SYNC_ON_CANCEL = 1,
217 ALWAYS_SYNC_ON_CANCEL = 2,
218 NUM_SYNC_ON_CANCEL_STATES
224 struct rw_semaphore cl_sem;
225 struct obd_uuid cl_target_uuid;
226 struct obd_import *cl_import; /* ptlrpc connection state */
228 /* max_mds_easize is purely a performance thing so we don't have to
229 * call obd_size_diskmd() all the time. */
230 int cl_default_mds_easize;
231 int cl_max_mds_easize;
232 int cl_default_mds_cookiesize;
233 int cl_max_mds_cookiesize;
235 enum lustre_sec_part cl_sp_me;
236 enum lustre_sec_part cl_sp_to;
237 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
239 /* the grant values are protected by loi_list_lock below */
240 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
241 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
242 unsigned long cl_dirty_transit; /* dirty synchronous */
243 unsigned long cl_avail_grant; /* bytes of credit for ost */
244 unsigned long cl_lost_grant; /* lost credits (trunc) */
246 /* since we allocate grant by blocks, we don't know how many grant will
247 * be used to add a page into cache. As a solution, we reserve maximum
248 * grant before trying to dirty a page and unreserve the rest.
249 * See osc_{reserve|unreserve}_grant for details. */
250 long cl_reserved_grant;
251 struct list_head cl_cache_waiters; /* waiting for cache/grant */
252 cfs_time_t cl_next_shrink_grant; /* jiffies */
253 struct list_head cl_grant_shrink_list; /* Timeout event list */
254 int cl_grant_shrink_interval; /* seconds */
256 /* A chunk is an optimal size used by osc_extent to determine
257 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
259 int cl_extent_tax; /* extent overhead, by bytes */
261 /* keep track of objects that have lois that contain pages which
262 * have been queued for async brw. this lock also protects the
263 * lists of osc_client_pages that hang off of the loi */
265 * ->cl_loi_list_lock protects consistency of
266 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
267 * ->ap_completion() call-backs are executed under this lock. As we
268 * cannot guarantee that these call-backs never block on all platforms
269 * (as a matter of fact they do block on Mac OS X), type of
270 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
271 * and blocking mutex on Mac OS X. (Alternative is to make this lock
272 * blocking everywhere, but we don't want to slow down fast-path of
273 * our main platform.)
275 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
276 * with client_obd_list_{un,}lock() and
277 * client_obd_list_lock_{init,done}() functions.
279 * NB by Jinshan: though field names are still _loi_, but actually
280 * osc_object{}s are in the list.
282 client_obd_lock_t cl_loi_list_lock;
283 struct list_head cl_loi_ready_list;
284 struct list_head cl_loi_hp_ready_list;
285 struct list_head cl_loi_write_list;
286 struct list_head cl_loi_read_list;
287 __u32 cl_r_in_flight;
288 __u32 cl_w_in_flight;
289 /* just a sum of the loi/lop pending numbers to be exported by /proc */
290 atomic_t cl_pending_w_pages;
291 atomic_t cl_pending_r_pages;
292 __u32 cl_max_pages_per_rpc;
293 __u32 cl_max_rpcs_in_flight;
294 struct obd_histogram cl_read_rpc_hist;
295 struct obd_histogram cl_write_rpc_hist;
296 struct obd_histogram cl_read_page_hist;
297 struct obd_histogram cl_write_page_hist;
298 struct obd_histogram cl_read_offset_hist;
299 struct obd_histogram cl_write_offset_hist;
301 /* lru for osc caching pages */
302 struct cl_client_cache *cl_cache;
303 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
304 atomic_t *cl_lru_left;
305 atomic_t cl_lru_busy;
306 atomic_t cl_lru_shrinkers;
307 atomic_t cl_lru_in_list;
308 struct list_head cl_lru_list; /* lru page list */
309 client_obd_lock_t cl_lru_list_lock; /* page list protector */
310 atomic_t cl_unstable_count;
312 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
313 atomic_t cl_destroy_in_flight;
314 wait_queue_head_t cl_destroy_waitq;
316 struct mdc_rpc_lock *cl_rpc_lock;
317 struct mdc_rpc_lock *cl_close_lock;
320 struct mutex cl_mgc_mutex;
321 struct local_oid_storage *cl_mgc_los;
322 struct dt_object *cl_mgc_configs_dir;
323 atomic_t cl_mgc_refcount;
324 struct obd_export *cl_mgc_mgsexp;
326 /* checksumming for data sent over the network */
327 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
328 /* supported checksum types that are worked out at connect time */
329 __u32 cl_supp_cksum_types;
330 /* checksum algorithm to be used */
331 cksum_type_t cl_cksum_type;
333 /* also protected by the poorly named _loi_list_lock lock above */
334 struct osc_async_rc cl_ar;
336 /* used by quotacheck when the servers are older than 2.4 */
337 int cl_qchk_stat; /* quotacheck stat of the peer */
338 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
339 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 53, 0)
340 #warning "please consider removing quotacheck compatibility code"
343 /* sequence manager */
344 struct lu_client_seq *cl_seq;
346 atomic_t cl_resends; /* resend count */
348 /* ptlrpc work for writeback in ptlrpcd context */
349 void *cl_writeback_work;
351 /* hash tables for osc_quota_info */
352 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
354 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
361 struct echo_client_obd {
362 struct obd_export *ec_exp; /* the local connection to osc/lov */
364 struct list_head ec_objects;
365 struct list_head ec_locks;
370 /* Generic subset of OSTs */
372 __u32 *op_array; /* array of index of
374 unsigned int op_count; /* number of OSTs in the array */
375 unsigned int op_size; /* allocated size of lp_array */
376 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
379 /* allow statfs data caching for 1 second */
380 #define OBD_STATFS_CACHE_SECONDS 1
382 struct lov_tgt_desc {
383 struct list_head ltd_kill;
384 struct obd_uuid ltd_uuid;
385 struct obd_device *ltd_obd;
386 struct obd_export *ltd_exp;
388 __u32 ltd_index; /* index in lov_obd->tgts */
389 unsigned long ltd_active:1,/* is this target up for requests */
390 ltd_activate:1,/* should target be activated */
391 ltd_reap:1; /* should this target be deleted */
395 struct lov_desc desc;
396 struct lov_tgt_desc **lov_tgts; /* sparse array */
397 struct ost_pool lov_packed; /* all OSTs in a packed
399 struct mutex lov_lock;
400 struct obd_connect_data lov_ocd;
401 struct proc_dir_entry *targets_proc_entry;
402 atomic_t lov_refcount;
403 __u32 lov_death_row; /* tgts scheduled to be deleted */
404 __u32 lov_tgt_size; /* size of tgts array */
407 cfs_hash_t *lov_pools_hash_body; /* used for key access */
408 struct list_head lov_pool_list; /* used for sequential access */
409 struct proc_dir_entry *lov_pool_proc_entry;
410 enum lustre_sec_part lov_sp_me;
412 /* Cached LRU and unstable data from upper layer */
415 struct rw_semaphore lov_notify_lock;
418 struct lmv_tgt_desc {
419 struct obd_uuid ltd_uuid;
420 struct obd_export *ltd_exp;
422 struct mutex ltd_fid_mutex;
423 unsigned long ltd_active:1; /* target up for requests */
426 enum placement_policy {
427 PLACEMENT_CHAR_POLICY = 0,
428 PLACEMENT_NID_POLICY = 1,
429 PLACEMENT_INVAL_POLICY = 2,
433 typedef enum placement_policy placement_policy_t;
437 struct lu_client_fld lmv_fld;
439 placement_policy_t lmv_placement;
440 struct lmv_desc desc;
441 struct obd_uuid cluuid;
442 struct obd_export *exp;
443 struct proc_dir_entry *targets_proc_entry;
445 struct mutex init_mutex;
450 int max_def_cookiesize;
453 int tgts_size; /* size of tgts array */
454 struct lmv_tgt_desc **tgts;
456 struct obd_connect_data conn_data;
459 struct niobuf_local {
460 __u64 lnb_file_offset;
461 __u32 lnb_page_offset;
464 struct page *lnb_page;
469 #define LUSTRE_FLD_NAME "fld"
470 #define LUSTRE_SEQ_NAME "seq"
472 #define LUSTRE_MDD_NAME "mdd"
473 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
474 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
475 #define LUSTRE_VVP_NAME "vvp"
476 #define LUSTRE_LMV_NAME "lmv"
477 #define LUSTRE_SLP_NAME "slp"
478 #define LUSTRE_LOD_NAME "lod"
479 #define LUSTRE_OSP_NAME "osp"
480 #define LUSTRE_LWP_NAME "lwp"
482 /* obd device type names */
483 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
484 #define LUSTRE_MDS_NAME "mds"
485 #define LUSTRE_MDT_NAME "mdt"
486 #define LUSTRE_MDC_NAME "mdc"
487 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
488 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
489 #define LUSTRE_OSC_NAME "osc"
490 #define LUSTRE_LOV_NAME "lov"
491 #define LUSTRE_MGS_NAME "mgs"
492 #define LUSTRE_MGC_NAME "mgc"
494 #define LUSTRE_ECHO_NAME "obdecho"
495 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
496 #define LUSTRE_QMT_NAME "qmt"
498 /* Constant obd names (post-rename) */
499 #define LUSTRE_MDS_OBDNAME "MDS"
500 #define LUSTRE_OSS_OBDNAME "OSS"
501 #define LUSTRE_MGS_OBDNAME "MGS"
502 #define LUSTRE_MGC_OBDNAME "MGC"
504 struct obd_trans_info {
506 /* Only used on the server side for tracking acks. */
507 struct oti_req_ack_lock {
508 struct lustre_handle lock;
512 struct llog_cookie oti_onecookie;
513 struct llog_cookie *oti_logcookies;
516 __u64 oti_pre_version;
520 * Events signalled through obd_notify() upcall-chain.
522 enum obd_notify_event {
525 /* Device connect start */
527 /* Device activated */
529 /* Device deactivated */
531 /* Device disconnected */
533 /* Connect data for import were changed */
536 OBD_NOTIFY_SYNC_NONBLOCK,
538 /* Configuration event */
540 /* Administratively deactivate/activate event */
541 OBD_NOTIFY_DEACTIVATE,
545 /* bit-mask flags for config events */
547 CONFIG_LOG = 0x1, /* finished processing config log */
548 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
549 CONFIG_TARGET = 0x4 /* one target is added */
553 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
554 * and liblustre being main examples).
556 struct obd_notify_upcall {
557 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
558 enum obd_notify_event ev, void *owner, void *data);
559 /* Opaque datum supplied by upper layer listener */
563 struct target_recovery_data {
564 svc_handler_t trd_recovery_handler;
565 pid_t trd_processing_task;
566 struct completion trd_starting;
567 struct completion trd_finishing;
570 struct obd_llog_group {
571 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
572 wait_queue_head_t olg_waitq;
574 struct mutex olg_cat_processing;
577 /* corresponds to one of the obd's */
578 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
581 struct obd_type *obd_type;
584 /* common and UUID name of this device */
585 char obd_name[MAX_OBD_NAME];
586 struct obd_uuid obd_uuid;
588 struct lu_device *obd_lu_dev;
590 /* bitfield modification is protected by obd_dev_lock */
592 obd_attached:1, /* finished attach */
593 obd_set_up:1, /* finished setup */
594 obd_recovering:1, /* there are recoverable clients */
595 obd_abort_recovery:1, /* recovery expired */
596 obd_version_recov:1, /* obd uses version checking */
597 obd_replayable:1, /* recovery is enabled;
599 obd_no_transno:1, /* no committed-transno notification */
600 obd_no_recov:1, /* fail instead of retry messages */
601 obd_stopping:1, /* started cleanup */
602 obd_starting:1, /* started setup */
603 obd_force:1, /* cleanup with > 0 obd refcount */
604 obd_fail:1, /* cleanup with failover */
605 obd_no_conn:1, /* deny new connections */
606 obd_inactive:1, /* device active/inactive
607 * (for /proc/status only!!) */
608 obd_no_ir:1, /* no imperative recovery. */
609 obd_process_conf:1, /* device is processing mgs config */
610 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
612 /* use separate field as it is set in interrupt to don't mess with
613 * protection of other bits using _bh lock */
614 unsigned long obd_recovery_expired:1;
615 /* uuid-export hash body */
616 cfs_hash_t *obd_uuid_hash;
617 /* nid-export hash body */
618 cfs_hash_t *obd_nid_hash;
620 cfs_hash_t *obd_nid_stats_hash;
621 struct list_head obd_nid_stats;
622 atomic_t obd_refcount;
623 struct list_head obd_exports;
624 struct list_head obd_unlinked_exports;
625 struct list_head obd_delayed_exports;
626 struct list_head obd_lwp_list;
628 spinlock_t obd_nid_lock;
629 struct ldlm_namespace *obd_namespace;
630 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
631 /* a spinlock is OK for what we do now, may need a semaphore later */
632 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
633 struct mutex obd_dev_mutex;
634 __u64 obd_last_committed;
635 spinlock_t obd_osfs_lock;
636 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
638 struct lvfs_run_ctxt obd_lvfs_ctxt;
639 struct obd_llog_group obd_olg; /* default llog group */
640 struct obd_device *obd_observer;
641 struct rw_semaphore obd_observer_link_sem;
642 struct obd_notify_upcall obd_upcall;
643 struct obd_export *obd_self_export;
644 struct obd_export *obd_lwp_export;
645 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
646 struct list_head obd_exports_timed;
647 time_t obd_eviction_timer; /* for ping evictor */
649 int obd_max_recoverable_clients;
650 atomic_t obd_connected_clients;
651 int obd_stale_clients;
652 /* this lock protects all recovery list_heads, timer and
653 * obd_next_recovery_transno value */
654 spinlock_t obd_recovery_task_lock;
655 __u64 obd_next_recovery_transno;
656 int obd_replayed_requests;
657 int obd_requests_queued_for_recovery;
658 wait_queue_head_t obd_next_transno_waitq;
659 /* protected by obd_recovery_task_lock */
660 struct timer_list obd_recovery_timer;
662 time_t obd_recovery_start;
663 /* seconds, for lprocfs_status */
664 time_t obd_recovery_end;
665 int obd_recovery_time_hard;
666 int obd_recovery_timeout;
667 int obd_recovery_ir_factor;
669 /* new recovery stuff from CMD2 */
670 struct target_recovery_data obd_recovery_data;
671 int obd_replayed_locks;
672 atomic_t obd_req_replay_clients;
673 atomic_t obd_lock_replay_clients;
674 /* all lists are protected by obd_recovery_task_lock */
675 struct list_head obd_req_replay_queue;
676 struct list_head obd_lock_replay_queue;
677 struct list_head obd_final_req_queue;
680 #ifdef HAVE_SERVER_SUPPORT
681 struct obd_device_target obt;
682 struct filter_obd filter;
684 struct echo_obd echo;
686 struct client_obd cli;
687 struct echo_client_obd echo_client;
691 /* Fields used by LProcFS */
692 unsigned int obd_cntr_base;
693 struct lprocfs_stats *obd_stats;
695 unsigned int obd_md_cntr_base;
696 struct lprocfs_stats *obd_md_stats;
698 struct proc_dir_entry *obd_proc_entry;
699 struct proc_dir_entry *obd_proc_exports_entry;
700 struct proc_dir_entry *obd_svc_procroot;
701 struct lprocfs_stats *obd_svc_stats;
702 struct lprocfs_seq_vars *obd_vars;
703 atomic_t obd_evict_inprogress;
704 wait_queue_head_t obd_evict_inprogress_waitq;
705 struct list_head obd_evict_list; /* protected with pet_lock */
708 * Ldlm pool part. Save last calculated SLV and Limit.
710 rwlock_t obd_pool_lock;
715 * A list of outstanding class_incref()'s against this obd. For
718 struct lu_ref obd_reference;
720 int obd_conn_inprogress;
723 enum obd_cleanup_stage {
724 /* Special case hack for MDS LOVs */
726 /* can be directly mapped to .ldto_device_fini() */
730 /* get/set_info keys */
731 #define KEY_ASYNC "async"
732 #define KEY_CAPA_KEY "capa_key"
733 #define KEY_CHANGELOG_CLEAR "changelog_clear"
734 #define KEY_FID2PATH "fid2path"
735 #define KEY_CHECKSUM "checksum"
736 #define KEY_CLEAR_FS "clear_fs"
737 #define KEY_CONN_DATA "conn_data"
738 #define KEY_EVICT_BY_NID "evict_by_nid"
739 #define KEY_FIEMAP "fiemap"
740 #define KEY_FLUSH_CTX "flush_ctx"
741 #define KEY_GRANT_SHRINK "grant_shrink"
742 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
743 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
744 #define KEY_INTERMDS "inter_mds"
745 #define KEY_LAST_ID "last_id"
746 #define KEY_LAST_FID "last_fid"
747 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
748 #define KEY_LOVDESC "lovdesc"
749 #define KEY_MAX_EASIZE "max_easize"
750 #define KEY_DEFAULT_EASIZE "default_easize"
751 #define KEY_MAX_COOKIESIZE "max_cookiesize"
752 #define KEY_DEFAULT_COOKIESIZE "default_cookiesize"
753 #define KEY_MDS_CONN "mds_conn"
754 #define KEY_MGSSEC "mgssec"
755 #define KEY_NEXT_ID "next_id"
756 #define KEY_READ_ONLY "read-only"
757 #define KEY_REGISTER_TARGET "register_target"
758 #define KEY_SET_FS "set_fs"
759 #define KEY_TGT_COUNT "tgt_count"
760 /* KEY_SET_INFO in lustre_idl.h */
761 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
762 #define KEY_CONNECT_FLAG "connect_flags"
764 #define KEY_CACHE_SET "cache_set"
765 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
766 #define KEY_OSP_CONNECTED "osp_connected"
770 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
771 #define IT_OPEN (1 << 0)
772 #define IT_CREAT (1 << 1)
773 #define IT_READDIR (1 << 2)
774 #define IT_GETATTR (1 << 3)
775 #define IT_LOOKUP (1 << 4)
776 #define IT_UNLINK (1 << 5)
777 #define IT_TRUNC (1 << 6)
778 #define IT_GETXATTR (1 << 7)
779 #define IT_EXEC (1 << 8)
780 #define IT_PIN (1 << 9)
781 #define IT_LAYOUT (1 << 10)
782 #define IT_QUOTA_DQACQ (1 << 11)
783 #define IT_QUOTA_CONN (1 << 12)
784 #define IT_SETXATTR (1 << 13)
786 static inline int it_to_lock_mode(struct lookup_intent *it)
788 /* CREAT needs to be tested before open (both could be set) */
789 if (it->it_op & IT_CREAT)
791 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
794 else if (it->it_op & IT_READDIR)
796 else if (it->it_op & IT_GETXATTR)
798 else if (it->it_op & IT_SETXATTR)
801 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
806 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
807 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
808 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
809 struct lu_fid op_fid4; /* to the operation locks. */
810 u32 op_mds; /* what mds server open will go to */
811 struct lustre_handle op_handle;
812 obd_time op_mod_time;
816 struct lmv_stripe_md *op_mea1;
817 struct lmv_stripe_md *op_mea2;
818 __u32 op_suppgids[2];
825 /* iattr fields and blocks. */
826 struct iattr op_attr;
828 loff_t op_attr_blocks;
830 /* Size-on-MDS epoch and flags. */
835 struct obd_capa *op_capa1;
836 struct obd_capa *op_capa2;
838 /* Various operation flags. */
839 enum mds_op_bias op_bias;
841 /* Used by readdir */
844 /* used to transfer info between the stacks of MD client
845 * see enum op_cli_flags */
848 /* File object data version for HSM release, on client */
849 __u64 op_data_version;
850 struct lustre_handle op_lease_handle;
853 #define op_stripe_offset op_ioepoch
854 #define op_max_pages op_valid
857 int (*md_blocking_ast)(struct ldlm_lock *lock,
858 struct ldlm_lock_desc *desc,
859 void *data, int flag);
862 struct md_enqueue_info;
863 /* metadata stat-ahead */
864 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
865 struct md_enqueue_info *minfo,
868 struct md_enqueue_info {
869 struct md_op_data mi_data;
870 struct lookup_intent mi_it;
871 struct lustre_handle mi_lockh;
872 struct inode *mi_dir;
873 md_enqueue_cb_t mi_cb;
878 struct module *o_owner;
879 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
880 void *karg, void *uarg);
881 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
882 __u32 keylen, void *key, __u32 *vallen, void *val,
883 struct lov_stripe_md *lsm);
884 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
885 __u32 keylen, void *key,
886 __u32 vallen, void *val,
887 struct ptlrpc_request_set *set);
888 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
889 int (*o_precleanup)(struct obd_device *dev,
890 enum obd_cleanup_stage cleanup_stage);
891 int (*o_cleanup)(struct obd_device *dev);
892 int (*o_process_config)(struct obd_device *dev, obd_count len,
894 int (*o_postrecov)(struct obd_device *dev);
895 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
897 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
898 /* connect to the target device with given connection
899 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
900 * granted by the target, which are guaranteed to be a subset of flags
901 * asked for. If @ocd == NULL, use default parameters. */
902 int (*o_connect)(const struct lu_env *env,
903 struct obd_export **exp, struct obd_device *src,
904 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
906 int (*o_reconnect)(const struct lu_env *env,
907 struct obd_export *exp, struct obd_device *src,
908 struct obd_uuid *cluuid,
909 struct obd_connect_data *ocd,
911 int (*o_disconnect)(struct obd_export *exp);
913 /* Initialize/finalize fids infrastructure. */
914 int (*o_fid_init)(struct obd_device *obd,
915 struct obd_export *exp, enum lu_cli_type type);
916 int (*o_fid_fini)(struct obd_device *obd);
918 /* Allocate new fid according to passed @hint. */
919 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
920 struct lu_fid *fid, struct md_op_data *op_data);
923 * Object with @fid is getting deleted, we may want to do something
926 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
927 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
928 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
929 __u64 max_age, struct ptlrpc_request_set *set);
930 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
931 struct lov_stripe_md *mem_src);
932 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
933 struct lov_mds_md *disk_src, int disk_len);
934 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
935 struct obdo *oa, struct lov_stripe_md **ea,
936 struct obd_trans_info *oti);
937 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
938 struct obdo *oa, struct lov_stripe_md *ea,
939 struct obd_trans_info *oti, struct obd_export *md_exp,
941 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
942 struct obd_info *oinfo, struct obd_trans_info *oti);
943 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
944 struct obd_trans_info *oti,
945 struct ptlrpc_request_set *rqset);
946 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
947 struct obd_info *oinfo);
948 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
949 struct ptlrpc_request_set *set);
950 int (*o_preprw)(const struct lu_env *env, int cmd,
951 struct obd_export *exp, struct obdo *oa, int objcount,
952 struct obd_ioobj *obj, struct niobuf_remote *remote,
953 int *nr_pages, struct niobuf_local *local,
954 struct obd_trans_info *oti, struct lustre_capa *capa);
955 int (*o_commitrw)(const struct lu_env *env, int cmd,
956 struct obd_export *exp, struct obdo *oa,
957 int objcount, struct obd_ioobj *obj,
958 struct niobuf_remote *remote, int pages,
959 struct niobuf_local *local,
960 struct obd_trans_info *oti, int rc);
961 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
962 ldlm_iterator_t it, void *data);
963 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
964 ldlm_iterator_t it, void *data);
965 int (*o_init_export)(struct obd_export *exp);
966 int (*o_destroy_export)(struct obd_export *exp);
968 /* llog related obd_methods */
969 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
970 struct obd_device *disk_obd, int *idx);
971 int (*o_llog_finish)(struct obd_device *obd, int count);
973 int (*o_import_event)(struct obd_device *, struct obd_import *,
974 enum obd_import_event);
976 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
977 enum obd_notify_event ev, void *data);
979 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
980 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
983 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
984 struct obd_quotactl *);
985 int (*o_quotactl)(struct obd_device *, struct obd_export *,
986 struct obd_quotactl *);
988 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
991 int (*o_pool_new)(struct obd_device *obd, char *poolname);
992 int (*o_pool_del)(struct obd_device *obd, char *poolname);
993 int (*o_pool_add)(struct obd_device *obd, char *poolname,
995 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
997 void (*o_getref)(struct obd_device *obd);
998 void (*o_putref)(struct obd_device *obd);
1000 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1001 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1002 * Also, add a wrapper function in include/linux/obd_class.h. */
1005 /* lmv structures */
1007 struct mdt_body *body;
1008 struct lov_stripe_md *lsm;
1009 struct lmv_stripe_md *lmv;
1010 #ifdef CONFIG_FS_POSIX_ACL
1011 struct posix_acl *posix_acl;
1013 struct mdt_remote_perm *remote_perm;
1014 struct obd_capa *mds_capa;
1015 struct obd_capa *oss_capa;
1018 struct md_open_data {
1019 struct obd_client_handle *mod_och;
1020 struct ptlrpc_request *mod_open_req;
1021 struct ptlrpc_request *mod_close_req;
1022 atomic_t mod_refcount;
1026 struct obd_client_handle {
1027 struct lustre_handle och_fh;
1028 struct lu_fid och_fid;
1029 struct md_open_data *och_mod;
1030 struct lustre_handle och_lease_handle; /* open lock for lease */
1035 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1037 struct lookup_intent;
1041 /* Every operation from MD_STATS_FIRST_OP up to and including
1042 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1043 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1044 * Operations after MD_STATS_LAST_OP are excluded from stats.
1045 * There are a few reasons for doing this: we prune the 17
1046 * counters which will be of minimal use in understanding
1047 * metadata utilization, we save memory by allocating 15
1048 * instead of 32 counters, we save cycles by not counting.
1050 * MD_STATS_FIRST_OP must be the first member of md_ops.
1052 #define MD_STATS_FIRST_OP m_close
1053 int (*m_close)(struct obd_export *, struct md_op_data *,
1054 struct md_open_data *, struct ptlrpc_request **);
1056 int (*m_create)(struct obd_export *, struct md_op_data *,
1057 const void *, int, int, __u32, __u32, cfs_cap_t,
1058 __u64, struct ptlrpc_request **);
1060 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1061 const union ldlm_policy_data *,
1062 struct lookup_intent *, struct md_op_data *,
1063 struct lustre_handle *, __u64);
1065 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1066 struct ptlrpc_request **);
1068 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1069 struct lookup_intent *,
1070 struct ptlrpc_request **,
1071 ldlm_blocking_callback, __u64);
1073 int (*m_link)(struct obd_export *, struct md_op_data *,
1074 struct ptlrpc_request **);
1076 int (*m_rename)(struct obd_export *, struct md_op_data *,
1077 const char *, int, const char *, int,
1078 struct ptlrpc_request **);
1080 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1081 int , void *, int, struct ptlrpc_request **,
1082 struct md_open_data **mod);
1084 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1085 struct obd_capa *, struct ptlrpc_request **);
1087 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1088 struct md_callback *cb_op, __u64 hash_offset,
1089 struct page **ppage);
1091 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1092 struct ptlrpc_request **);
1094 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1095 struct obd_capa *, obd_valid, const char *,
1096 const char *, int, int, int, __u32,
1097 struct ptlrpc_request **);
1099 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1100 struct obd_capa *, obd_valid, const char *,
1101 const char *, int, int, int,
1102 struct ptlrpc_request **);
1104 int (*m_intent_getattr_async)(struct obd_export *,
1105 struct md_enqueue_info *,
1106 struct ldlm_enqueue_info *);
1108 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1109 struct lu_fid *, __u64 *bits);
1111 #define MD_STATS_LAST_OP m_revalidate_lock
1113 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1114 struct obd_capa **);
1116 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1118 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1119 ldlm_iterator_t, void *);
1121 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1122 struct md_open_data *);
1124 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1125 struct ptlrpc_request **);
1127 int (*m_init_ea_size)(struct obd_export *, int, int, int, int);
1129 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1130 struct obd_export *, struct obd_export *,
1131 struct lustre_md *);
1133 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1135 int (*m_merge_attr)(struct obd_export *,
1136 const struct lmv_stripe_md *lsm,
1137 struct cl_attr *attr);
1139 int (*m_update_lsm_md)(struct obd_export *, struct lmv_stripe_md *lsm,
1140 struct mdt_body *, ldlm_blocking_callback);
1142 int (*m_set_open_replay_data)(struct obd_export *,
1143 struct obd_client_handle *,
1144 struct lookup_intent *);
1146 int (*m_clear_open_replay_data)(struct obd_export *,
1147 struct obd_client_handle *);
1149 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1151 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1152 const struct lu_fid *, ldlm_type_t,
1153 ldlm_policy_data_t *, ldlm_mode_t,
1154 struct lustre_handle *);
1156 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1157 ldlm_policy_data_t *, ldlm_mode_t,
1158 ldlm_cancel_flags_t flags, void *opaque);
1160 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1161 renew_capa_cb_t cb);
1163 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1164 const struct req_msg_field *, struct obd_capa **);
1166 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1167 struct obd_capa *, __u32,
1168 struct ptlrpc_request **);
1170 int (*m_get_fid_from_lsm)(struct obd_export *,
1171 const struct lmv_stripe_md *,
1172 const char *name, int namelen,
1173 struct lu_fid *fid);
1176 struct lsm_operations {
1177 void (*lsm_free)(struct lov_stripe_md *);
1178 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1179 struct obd_export *md_exp);
1180 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1182 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1184 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1185 __u16 *stripe_count);
1186 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1187 struct lov_mds_md *lmm);
1190 extern const struct lsm_operations lsm_v1_ops;
1191 extern const struct lsm_operations lsm_v3_ops;
1192 static inline const struct lsm_operations *lsm_op_find(int magic)
1200 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1205 static inline struct md_open_data *obd_mod_alloc(void)
1207 struct md_open_data *mod;
1211 atomic_set(&mod->mod_refcount, 1);
1215 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1216 #define obd_mod_put(mod) \
1218 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1219 if ((mod)->mod_open_req) \
1220 ptlrpc_req_finished((mod)->mod_open_req); \
1221 OBD_FREE_PTR(mod); \
1225 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1226 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1228 /* return 1 if client should be resend request */
1229 static inline int client_should_resend(int resend, struct client_obd *cli)
1231 return atomic_read(&cli->cl_resends) ?
1232 atomic_read(&cli->cl_resends) > resend : 1;
1236 * Return device name for this device
1238 * XXX: lu_device is declared before obd_device, while a pointer pointing
1239 * back to obd_device in lu_device, so this helper function defines here
1240 * instead of in lu_object.h
1242 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1244 return lu_dev->ld_obd->obd_name;
1247 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1252 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1255 /* caller does not care of idx */
1259 /* volatile file, the MDT can be set from name */
1260 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1261 /* if no MDT is specified, use std way */
1262 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1264 /* test for no MDT idx case */
1265 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1266 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1270 /* we have an idx, read it */
1271 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1272 *idx = strtoul(start, &end, 16);
1274 * no digit, no trailing :, negative value
1276 if (((*idx == 0) && (end == start)) ||
1277 (*end != ':') || (*idx < 0))
1282 /* bad format of mdt idx, we cannot return an error
1283 * to caller so we use hash algo */
1284 CERROR("Bad volatile file name format: %s\n",
1285 name + LUSTRE_VOLATILE_HDR_LEN);
1289 static inline int cli_brw_size(struct obd_device *obd)
1291 LASSERT(obd != NULL);
1292 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1295 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1296 * of the client should be increased accordingly to avoid sending fragmented
1297 * RPCs over the network when the client runs out of the maximum dirty space
1298 * when so many RPCs are being generated.
1300 static inline void client_adjust_max_dirty(struct client_obd *cli)
1303 if (cli->cl_dirty_max_pages <= 0)
1304 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1305 >> PAGE_CACHE_SHIFT;
1307 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1308 cli->cl_max_pages_per_rpc;
1310 if (dirty_max > cli->cl_dirty_max_pages)
1311 cli->cl_dirty_max_pages = dirty_max;
1314 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1315 cli->cl_dirty_max_pages = totalram_pages / 8;
1318 #endif /* __OBD_H */